CCSA Conference on data quality for international organizations Rome, 7-8 July Some quality issues in imputation: ILO experiences A.S. Young ILO Bureau of Statistics

CCSA Conference on data quality for international organizations Rome, 7-8 July Quality issues Quality as “Fit for purpose” Dimensions examined: –t–the purpose –t–the approach taken in making the imputations; and –t–the dissemination policies used.

CCSA Conference on data quality for international organizations Rome, 7-8 July Purpose Evolution over time of a characteristic – national level Evolution over time of the distribution of a characteristic over countries – international level Issue 1: Given these different purposes, should imputed values derived for a country at a point in time through the two processes be expected to be the same?

CCSA Conference on data quality for international organizations Rome, 7-8 July Issue 2:Implicit vs Explicit Holt’s proposal “Suggestion 6: Agencies should seek to establish explicit imputation methods where thorough empirical analyses can demonstrate that these are robust and methodologically sound.”

CCSA Conference on data quality for international organizations Rome, 7-8 July Approach Issue 3: Is an analysis of ‘missingness’ important? Model selection: goodness-of-fit and predictive power Diagnostics: –T–Tabassum and Holt (2004), “The model fitting process must not be regarded as automatic and considerable checking and validation of results is required …”

CCSA Conference on data quality for international organizations Rome, 7-8 July

CCSA Conference on data quality for international organizations Rome, 7-8 July Implausible values Issue 4: What to do with implausible imputed values? Abayomi et al: “When problems are found, the imputer should refine the imputation model to create improved imputations that are consistent.” Issue 5: Should the functional model be the same for all regions, if separate regional imputations are done?

CCSA Conference on data quality for international organizations Rome, 7-8 July

CCSA Conference on data quality for international organizations Rome, 7-8 July Uses and dissemination Handling of imputation (plausibility) errors? Handling of revised values? Consultation with countries: quality assurance of methods used Acceptance by users: Full documentation of process used